Just specific to Nemenyi and Dunns tests, I didn't check the other parts of
this discussion.
They were discussed here
https://github.com/statsmodels/statsmodels/issues/852 (starting after a
few comments)
with code available in gists but not yet in a PR for statsmodels
Josef
On Sat, Oct 31, 2015
Thank you all,
The Orange will work just fine:
http://docs.orange.biolab.si/3/modules/evaluation.cd.html
Andreas, I'm not sure if this is the kind of thing that needed
to be in sklearn, maybe scipy (stats) and matplotlib (graph).
Thanks,
On Thu, Oct 29, 2015 at 1:55 PM, Arnaud Joly wrote:
> s
scipy allows to perform the friedman test.
Orange has the tool to drawn the critical distance diagram.
And you can easily compute the critical distance using stats model:
from statsmodels.stats.libqsturng import qsturng
q_alpha = qsturng(1 - alpha, n_methods, np.inf) / np.sqrt(2)
cd = q_alpha * n
Sorry, don't know of a package. But it might be interesting for sklearn?
So that's a Nemenyi test?
https://en.wikipedia.org/wiki/Nemenyi_test
I never heard of that but it sounds interesting.
It seems a bit hard to interpret, though.
Also: does the diagram punt if the initial multiple comparison
Hi,
Do you guys know any tool to generate CDdiagram - in order to evaluate the
difference of performance of sklearn classifiers?
http://theoval.cmp.uea.ac.uk/matlab/critdiff/cd1.png​
There is a R package called performanceEstimation which has
a CDdiagram implementation, but it uses an specific
